Player not loading? Watch on YouTube
This tutorial walks through Ollama's preliminary MLX support on Apple Silicon using Msty Studio's Local AI Services. The supplied description specifies Msty Studio 2.6.2 and Local AI Service 0.19.0 or newer. The speaker cites a recommendation of at least 32 GB of unified memory for the demonstrated model, qwen3.5:35b-a3b-coding-nvfp4.
The setup starts with version checks in settings, then moves to the model hub's local AI section and Ollama model browser. Because the model listing was not yet available during recording, the speaker demonstrates entering the exact model name to download it from Ollama's registry. After installation, they rename its display label and select it in a new conversation to run models locally.
The demonstration also covers thinking controls. The Ollama-backed model exposes default and none, while the speaker says Msty's separate MLX bundle has higher and lower thinking levels too. A greeting produces a response that the speaker describes as quick, but the video provides no measured performance comparison.
Finally, the model generates code for a Python game. The speaker does not execute or test that code. They also state that Msty Studio's agent mode does not yet support local models at the time of recording.
